Building AI that runs itself. Senior Engineer at @TradeRepublic. Speaker. iOS, Swift, Rust and Python. Making https://xitter-newx.xyz-mikolabs.workers.dev/t.co/Kq7FisXWls - https://xitter-newx.xyz-mikolabs.workers.dev/t.co/IJLnfrldRk
Berlin
Joined April 2009
- Tweets1.4K
- Following600
- Followers775
- Likes989
Scanning slow query logs for missing indexes is a good fit for an agent, since an EXPLAIN shows right away if it got it right. On write-heavy tables I have a feeling its suggestions will still need a human look before they ship.
1
1
10
AI agents get reviewed against a flawless employee, while the person doing that job today is tired, distracted and rushing. I wouldn't be surprised if a lot of useful automation stalls on that comparison alone.
1
1
29
Interactive AI story series where viewers vote on the next episode can be produced by almost anyone now with Claude, so my bet is the feeds fill up fast and only the ones with a real storyteller behind them keep an audience.
11
Rewarding subscribers for how long they've stayed is rare in software. The discounts usually go to new customers and the long-time ones get the price increase.
My guess is more AI tools start doing it, since switching between them is so easy now.
11
Agent-written code keeps repeating the same kinds of mistakes, so I'm pretty sure checks that catch a whole class of bugs pay off more than reviewing every diff line by line.
9
Local or API stops being a choice once the harness is mine.
Small local models take the routine calls and a rented frontier model only gets the hard ones, and I suspect that routing is where most of the savings come from.
8
Most of the work in an AI-made commercial seems to be choosing. Hundreds of keyframes get generated to keep a couple dozen shots, and I'd expect that selection, plus keeping the look consistent from shot to shot, to stay the human part for a while.
13
An AI adoption strategy is mostly a big company problem, since a small team tries the tool and drops it the same week if it doesn't help.
Event photos used to be the proof that the dinner happened, and I'm not sure they stay that once people start rendering the ones they forgot to take.
17
Once people already pay for a chatbot that does a rough version of it, a monthly subscription for a simple utility is a hard sell.
14
Managing agents is mostly writing clear specs and reviewing pull requests, which senior engineers already do every day, so I wouldn't be surprised if they end up being the people companies hire for it.
1
1
34
Dark patterns in cancel flows lose most of their power when an agent is the one clicking through them, I'd bet.
16
Fewer native mobile and frontend openings, yet apps now need to impress more than ever. My guess is companies hand that work to a few senior engineers who own the whole product, with LLMs covering the simple screens.
16
Now that customers ask ChatGPT before they ever visit a site, a plain text FAQ might be worth more than any video made from it.
14
Getting a first $100 online in a day got harder with AI tools, I think, not easier. Everyone has the same tools and ships the same ideas on day one, so the money still goes to whoever already has people willing to pay them.
1
13
ZONOS3-Mini cloning voices on the user's own GPU in a browser tab, under Apache 2.0, makes per-character TTS pricing a hard sell for small apps.
20
Web performance is a good fit for agents. The numbers are easy to measure and the profiling work is tedious, which is why it kept losing to feature work.
I'd expect plenty of apps to get faster without anyone on the team ever prioritizing it.
16
Error budgets were sized for teams shipping a few deploys a day, and I doubt they hold up once agents ship dozens.
17
A reset people can save for their heaviest day is a nice touch. Still, the bigger fix to me would be showing how much of the limit each request used, so bugs like these get noticed by users and not only by the team.
We've given all Pro, Max, Ultra, and Team users one free full reset, across all models and limits. Use it anytime before December 10 at 2:00PM PT.
We recently fixed a few small issues that used more limits than intended, and the Fable resets didn't do much for those of you who'd already hit your weekly limits for all models. While these issues didn't affect everyone, we want to make up for the trouble.
You'll find it in Settings > Usage, or at clau.de/reset.
17